Spectral Properties of General Neurofunctions
Abstract
The article deals with the study of the spectral properties of generalized neurofunctions. Generalized neural elements are considered and conditions of realization of Boolean functions on these elements are investigated. Enhanced functionality of generalized neural elements makes it possible to develop effective methods for coding, compression, discrete signal recognition, and reducing the number of elements in neural circuits that are intended to solve problems in the field of prediction, artificial intelligence, medicine, and more. The concept of generalized Boolean neurofunction and the characteristic vector of the function of logic algebra with respect to a given system of characters are introduced. A characteristic vector of a Boolean function with respect to a given character system is constructed from the corresponding spectral coefficients of this function in the Walsh–Adamar basis system.
The spectral properties of functions of logic algebra realized by one generalized neural element are investigated. Using the properties of characteristic vectors of Boolean functions, the criterion for their realization by one generalized neural element was obtained. From the criteria given in the paper, it follows directly that Boolean functions that are realized by a single generalized neural element are uniquely determined by their characteristic vectors with respect to a given character system. If the character system in respect of which the UNE is considered contains elements, then for one-sided determination of the Boolean function from the arguments, which is realized by one such generalized neural element, enough spectral coefficients from the spectral decomposition of this function in the Walsh–Adamar basis function system. The results obtained can be effectively used for the compression of generalized Boolean neurofunctions, as well as for the development of synthesis methods for a generalized neural element.
